From Basalts to Badlands
Modelling long-term landscape response to lava
damming of an upland catchment in western Turkey
Prof. Dr A. Veldkamp Professor of Land Dynamics Wageningen University
Dr J.M. Schoorl
Assistant professor, Soil Geography and Landscape Group Wageningen University
Dr A.J.A.M. Temme
Assistant professor, Soil Geography and Landscape Group Wageningen University
Prof. Dr C.J. Ritsema, Wageningen University Prof. Dr V.G. Jetten, University of Twente, Enschede
Prof. Dr J. Wainwright, University of Durham, United Kingdom Dr A. Mather, University of Plymouth, United Kingdom
This research was conducted under the auspices of the C.T. De Wit Graduate School for Production Ecology and Resource Conservation (PE&RC)
From Basalts to Badlands
Modelling long-term landscape response to lava
damming of an upland catchment in western Turkey
Wouter van Gorp
submitted in fulfilment of the requirements for the degree of doctor at Wageningen University
by the authority of the Rector Magnificus Prof. Dr M.J. Kropff,
in the presence of the
Thesis Committee appointed by the Academic Board to be defended in public
on Friday 12 September 2014 at 11 a.m. in the Aula.
From basalts to badlands. Modelling long-term landscape response to lava damming of an upland catchment in western Turkey,
PhD thesis, Wageningen University, Wageningen, NL (2014) With references, with summaries in Turkish, Dutch and English ISBN: 978-94-6257-048-1
Hans van Gorp Hankie van Gorp - Nieuwenhuis
Chapter 1 General introduction . . . 9
1.1 Introduction . . . 9
1.2 Field and dating based landscape reconstruction . . . 10
1.3 Quaternary landscape reconstruction of Mediterranean catchments . . . 11
1.4 Landscape evolution modelling . . . 13
1.5 Study area . . . 15
1.6 Aim and research questions . . . 16
1.7 Thesis outline . . . 17
Chapter 2 Landscape evolution modelling of naturally dammed rivers. . . 19
2.1 Introduction . . . 20
2.2 Methods . . . 21
2.3 Results . . . 27
2.4 Discussion . . . 36
2.5 Conclusion . . . 42
Chapter 3 Fluvial response to Holocene volcanic damming and breaching in the Gediz and Geren rivers, western Turkey . . . 45
3.1 Introduction . . . 46
3.2 Setting . . . 47
3.3 Methods . . . 52
3.4 Results and discussion . . . 57
3.5 Conclusion . . . 73
Appendix A3.1. 40Ar/39Ar dating results . . . 74
Appendix A3.2. Luminescence dating . . . 78
Chapter 4 Modelling long-term (300 ka) upland catchment response to multiple lava damming events . . . 85 4.1 Introduction . . . 86 4.2 Study site . . . 87 4.3 Methods . . . 88 4.4 Results . . . 95 4.5 Discussion . . . 103 4.6 Conclusion . . . 105
Chapter 5 Long-term response of the Geren Catchment (western Turkey) to lava dam influenced base level change . . . 107
5.1 Introduction . . . 108
5.2 Regional setting . . . 109
5.3 Materials and methods . . . 110
5.4 Fieldwork and dating results . . . 113
5.5 Discussion . . . 120
5.6 Conclusion . . . 131
Appendix A5.1. Radiocarbon calibration . . . 132
Appendix A5.2. 40Ar/39Ar Age estimates and geochemistry . . . 133
6.1 Drivers of middle Pleistocene to Holocene evolution of the Geren Catchment . . 147
6.2 LAPSUS, an ever evolving reduced complexity landscape evolution model . . . 154
6.3 Research implications . . . 156 6.4 Final conclusions . . . 158 References . . . 161 Summary . . . 169 Samenvatting . . . 171 Özet . . . 174 Dankwoord . . . 177 List of Publications . . . 179
The face of a landscape is the legacy of its history. Its current shape reflects the formation history of the relief, rocks and soil it is composed of. It reflects the tectonic, climatic and, in recent geological time, the human impacts it underwent. For instance, upland and tectonically uplifted landscapes are mostly erosive, while subsiding or lowland areas, mostly reflect its burial by sediments delivered from upstream. However, deposition occurs in erosion-dominated landscapes as well, and vice versa. Specific landscapes within each of these regions still differ due to the rate and magnitude of tectonic, climatic and human history. Therefore, present and future landscape evolution is not only determined by climatic, tectonic and human driving mechanisms, it is also determined by its past state and properties, and by its specific historical pathway (Phillips, 2006).
Unfortunately for those who want to understand present or even predict future landscape evolution, many traces of a landscape’s past evolution are erased during the process, making reconstruction of a past landscape like “solving a large puzzle of which only few pieces are left”. Nevertheless, key pieces of this puzzle derived from field-based reconstructions can tell geomorphologists a lot about this old landscape and the processes that shaped it.
In geomorphology we study the shape and other properties of the landscape combined with the processes that shaped the landscape. This helps us answering questions about why the landscape looks like it does, how it was formed and how it will further evolve. For example, in fluvial geomorphology, the grade of a river stretch can conceptually be determined by the balance of water discharge and sediment load the water can carry (Gilbert, 1880). Thus, when the slope of the river declines, stream capacity lowers and leading to decreased erosion or increased sediment deposition. When the slope increases, stream capacity increases, leading to increased erosion or decreased deposition. This concept was supposedly applicable to all rivers regardless of their location or state, illustrating that specific river evolution could be described by, or deduced from, a general law irrespective of its specific location.
However, landforms are shaped not only as a function of their geological structure and processes, but also with respect to the time that these processes had to shape the land, as classically postulated by Davis (1899). In his geographical cycle, a landscape becomes uplifted and then evolves through several phases of maturity, before it may be uplifted again. Young landscapes show initially incised plateaus, mature landscapes show an incised river system with slopes adjusting to it, while old landscapes have all their relief removed to become relatively flat. These classic examples (of Gilbert and Davis) were for a long time the most influential geomorphological models, although there would be many to follow (e.g. Grant et al., 2013). They both bear some truth, but in the end are incomplete in their explanation of landscape evolution. However, they hypothesized how a landscape works from (field) observations, using inductive reasoning, a tool that remains valuable today when doing explorative fieldwork. The concept of the graded river was put forward by Mackin (1948).
This concept related the river profile to its base level, which acts as a downstream control on profile evolution.
With the increasing availability of information on past climate variability, from marine sediment records, ice cores and lacustrine records (e.g. Emiliani and Milliman, 1966; Lisiecki and Raymo, 2005; Tzedakis et al., 2006), the notion of climate change as an important driver in long-term landscape evolution became apparent. In addition, it became acknowledged that landscapes that are influenced by a constant driver (e.g. constant rainfall, constant uplift rate), display features like thresholds and complex response (Schumm, 1973). This non-linear response of landscapes to both constant and changing drivers has been the subject of many studies since. Another concept that relates to this is equifinality: similar landscapes can arise as a result of different processes and pathways. For instance, alternating aggradation-incision phases of a river system can be caused by climatic variation or base level change. To investigate which pathway led to the current landscape, field-based reconstruction can be used.
1.2 Field and dating based landscape reconstruction
1.2.1 Sedimentary archives: key pieces of the puzzle
In deposition-dominated environments such as marine basins, lakes, deltas, subsiding grabens and foreland basins, a stacked sedimentary record can reveal its depositional history. Texture (clay, silt, sand, gravel) can inform us on the available transport energy during which the sediment was deposited (low to high energy) or weathering conditions. The presence of fossils or pollen can inform us on palaeoclimatic conditions, while organic rich layers in deep water sediments can indicate climatic variation, and fine laminated lake varves show annual deposition phases. In erosion-dominated landscapes, such as uplifted plateaus, horsts, or mountain ranges, it is the preservation of certain landforms or depositional events in and along its valleys that inform us on its landscape history. For instance, under ongoing incision, remnants of an old river valley elevated above the current river level indicate its past valley floor or floodplain. Preserved sorted stratified sands and gravels indicate its former fluvial activity, while poorly sorted and structured sediments indicate debris flows or landslides. Calcium carbonate cementation or the presence of an organic rich topsoil within these deposits can show subsequent stable landscape and climatic conditions, while truncated soils or soils buried by slope or fluvial deposits indicate subsequent landscape instability. In time, erosion-dominated landscapes could have changed into deposition-dominated landscapes and vice-versa. For example, finding dissected lake sediments in an erosional landscape indicate major shift from depositional to erosional conditions, for example due to a dam breach event. Interpreting sedimentary records leads to a stratigraphy. Dating techniques can time these ancient deposits and landscape surfaces. This can link them to climatic periods and for instance allows calculation of incision rates.
1.2.2 Dating as a tool for landscape reconstruction
Dating possibilities of Quaternary materials and surfaces have increased rapidly. Methods used in this thesis are radiocarbon dating, feldspar luminescence dating and 40Ar/39Ar-dating. Radiocarbon dating of the time since death of organic material is at present possible until
~50 ka (Reimer et al., 2013). Luminescence dating is used to date the burial age of quartz or feldspars in sandy deposits, by measuring the amount of natural radiation the grains received from surrounding materials since they were last exposed to light. Feldspar luminescence dating can be applied to a range of ages, from young Holocene samples (Reimann and Tsukamoto, 2012) to ~400 ka (Buylaert et al., 2012), depending on dose rate (see Chapter 3 and 5 for further explanation). 40Ar/39Ar-dating can be used to date lava flows, ranging from young Holocene (Wijbrans et al., 2011) to no maximum, by measuring the amount of potassium-derived radiogenic argon built up since cooling time in a stepped-heating procedure (McDougall and Harrison, 1999). If lava flows filled river valleys, this estimates the time of potential damming of this valley and serves as a minimum age for underlying fluvial deposits (Maddy et al., submitted). Other commonly used dating techniques include cosmogenic nuclide dating, which can date the time since a surface has been exposed and its erosion rate (Gosse and Phillips, 2001) and ranges from 0.1 ka to > 1 Ma (Darvill, 2013).
Ages of fluvial terrace abandonment using 10Be have been determined up to ~700 ka (Rixhon
et al., 2011; Viveen et al., 2012). U-series dating, which has a ~600 ka age range (Cheng et al., 2000) is used to date carbonate precipitation and can be used to date coral, travertines (e.g. Veldkamp and Kroonenberg, 1993) or calcretes (e.g. Candy and Black, 2009).
Age estimates have uncertainties at different levels. First, they have an analytical uncertainty. This can be quantified and is dependent on measurement precision and sample quality. Second, based on sample characteristics, an informed choice is made on which type of age model or statistic to use to calculate the age. For instance in luminescence dating one could look at either the Central Age Model or the Minimum Age Model, which is an informed decision based on the expectation that grains are either generally well-bleached, or partially well-bleached, given that no post-burial disturbance occurred (Galbraith et al., 1999). In 40Ar/39Ar-dating the total fusion age is the age derived from all measured radiogenic argon. However, the stepped heating procedure can reveal that part of the sample generates e.g. an older age, implying contamination by alteration or inheritance of radiogenic argon from not entirely molten crystals in the magma (McDougall and Harrison, 1999). If several consecutive steps which give statistically the same age are observed, an acceptable plateau age can be derived. These choices are usually driven by statistics of the date and can therefore be seen as well informed decisions. However, sample context and interpretation have to be well-conveyed, as uncertainties in interpretation add another uncertainty on top of these analytical uncertainties. Using multiple dating techniques in one study can be an advantage to cross-validate age results. Furthermore, they can complement each other in Quaternary landscape reconstruction studies and allow correlations to other areas.
1.3 Quaternary landscape reconstruction of Mediterranean catchments
The catchment under study in this thesis is located in the eastern Mediterranean. The catchments surrounding the Mediterranean have been formed as a result of the Cenozoic collision of the African plate and the Eurasian plate. The complex tectonic processes and structures resulting from this collision caused areas of uplift and subsidence. All Mediterranean basins have the legacy of the Messinian Salinity crisis, causing a huge drop in base level and triggering all systems to incise significantly during the Late Miocene. This was
followed by a large transgression during the Early Pliocene, burying the Messinian gorges and related landscapes (Schoorl and Veldkamp, 2003). Uplifted areas with fluvial terraces around the Mediterranean have been the subject of many studies, because their sedimentary record and palaeo reconstruction can provide insight in fluvial development since their initiation (Maddy et al., 2007), leading to increased understanding of how these systems respond to climatic and tectonic constraints.
1.3.1 Climate and tectonic drivers
Reconstruction of Mediterranean catchments has been focussed on correlation of Quaternary river terrace staircases of large and regional rivers. These staircases occur in many uplifting areas in the world and they record a climatically controlled incision history, suggesting a global uplift – incision activity (Bridgland and Westaway, 2008; Gibbard and Lewin, 2009). Additionally, Mediterranean-wide aggradation sequences of catchments of different size and tectonic setting are suggested to be correlated to Pleistocene climatic fluctuations (Macklin et al., 2002). These studies thus focus on global similarities of regionally uplifted areas, all showing a similar climatically driven fluvial incision response.
1.3.2 Local drivers
Besides research on terrace staircase correlation, studies have focussed on specific catchment response of uplifting river reaches or basins. These studies illustrated how the specific base level history and boundary conditions of catchments influenced their Quaternary evolution. For instance in tectonically active southeast Spain, catchment response occurred through river capture-driven accelerated incision waves which still migrate through the catchment at present (Mather et al., 2002). In another catchment in southeast Spain, tectonic tilting invoked headward erosion and drainage expansion (Stokes and Mather, 2003). Examples of specific catchment response to uplift include northwest Spain, where local tectonics influence local terrace formation and preservation (Viveen et al., 2013). A study in northwest Europe indicates diachronous terrace abandonment and knickpoint retreat of a Meuse tributary as a response to uplift (Demoulin et al., 2012).
Other studies indicate the response to damming events. For instance in the United States, where Quaternary evolution of the Owyhee River has been seriously influenced by multiple lava dams (Ely et al., 2012). A catchment in southeast Spain has been temporally dammed in the late Pleistocene causing disequilibrium conditions in the Holocene (Baartman et al., 2011). In the Upper Gediz River area in Turkey (the study area of this thesis, see section 1.5), a tributary catchment endured major drainage diversion due to Early Pleistocene lava damming, while the trunk river was only mildly disturbed. Thus, local drivers such as local tectonics, river capture and damming events can significantly influence long-term catchment evolution and these studies suggest that knowledge of local history of catchments is important before making any regional correlations to climatic or uplift curves. This does not only apply to smaller catchments, but to regional rivers such as the Gediz River as well.
1.3.3 Natural dams
significant (Costa and Schuster, 1988). Natural dams obstruct or divert water and sediment routing and create local base levels within river reaches (Burchsted et al., 2014). They can be formed due to biological activity, such as beaver dams (Levine and Meyer, 2014) or log jams (Wohl and Beckman, 2014), biochemical and hydrochemical systems such as travertine and tufa dams (Florsheim et al., 2013; Ordóñez et al., 2005; Özkul et al., 2014), ice, moraine and landslide dams (Korup and Tweed, 2007), volcanic edifices (Macaire et al., 1992), volcanic debris flows (Capra, 2007) and lava dams (Ely et al., 2012; Hamblin, 1990). Landslide dams are often short-lived (Ermini and Casagli, 2003) due to their composition of often unconsolidated materials. Dam longevity however, depends on their composition Nevertheless, landslide and lava dams which have been stable for a longer duration occur and can impact river channel evolution at a 104 timescale (Ely et al., 2012; Korup et al., 2010). Natural dams often breach catastrophically, creating outburst floods which can exceed those of peak meteoric floods, resulting in significant geomorphic impact on the landscape (O’Connor et al., 2013). However, long-lived lava dams are reported to have been gradually removed (Ely et al., 2012) and long-term knickpoint persistence due to landslide damming and breaching occurs (e.g. Korup, 2013).
1.4 Landscape evolution modelling
In the second half of the past century, many researchers started quantifying “geomorphic transport laws”, which are mathematical expressions of physical principles that describe erosion or movement of material due to certain processes (e.g. Dietrich et al., 2003), and which can be used to deduce landscape change at a specific area. It was the start of the digital era that brought application and testing hypotheses and equifinality issues within reach. In the last three decennia, increased computer power resulted in the development of spatial simulation models, such as braided river models (Murray and Paola, 1994) or models of Aeolian dunes (Werner, 1995). In addition, the increased availability of high-quality Digital Elevation Models (DEMs, a digital representation of the landscape’s surface elevation) led to the development of Landscape Evolution Models (LEMs) (Tucker and Hancock, 2010). LEMs are computer models that simulate landscape change over time. Almost without exception, these models use simplified process-descriptions that do not do full justice to Newtonian physics. However, their outcomes often show sometimes unexpected complex features which can help conceptualizing an actual field situation. Examples are lagged response to an external driver (Temme and Veldkamp, 2009), dampening of external signals (Veldkamp and Tebbens, 2001), self-organised criticality of catchments (Van De Wiel and Coulthard, 2010) and spatially and topography driven complex erosion-sedimentation dynamics (Schoorl et al., 2014). Many LEMs use rectangular-gridded DEMs as an input, such as LEMs SIBERIA (Hancock et al., 2010) and CEASAR (Coulthard et al., 2005) and the LEM used in this thesis, LAPSUS (Schoorl et al., 2000; Schoorl et al., 2002). More elaborate reviews of LEMs can be found in Chapter 2 and 4.
1.4.1 LAPSUS landscape evolution model
The LAPSUS LEM was originally designed as a spatially explicit water runoff and erosion-deposition model (Schoorl et al., 2000; Schoorl et al., 2002). It is able to route water down using a multiple flow routine (Freeman, 1991; Quinn et al., 1991), in which water can be
divided over multiple downstream cells. To calculate a cell’s capacity to erode, it uses the continuity equation, based on principles of Kirkby (1971), where potential sediment transport, or transport capacity, is depending on the product of waterflow to a power and slope to a power. This principle, and elaborations of this principle still underlie many long-term landscape evolution studies (e.g. Lague, 2014). Calculation of actual sediment in transport in LAPSUS is based on principles of (Foster and Meyer, 1972; Foster and Meyer, 1975), in which actual runoff erosion or deposition of a location depends on the amount of sediment already in transport compared with transport capacity. The resulting sediment available for erosion or deposition will then be partly eroded or deposited, depending on an erodibility or sedimentation factor. In LAPSUS, the use of this relation was extended to landscapes, demonstrating the importance of spatial erosion deposition patterns (Schoorl et al., 2002). It has since then been extended to be able to deal with processes such as tillage erosion (Schoorl et al., 2004), shallow landsliding (Claessens et al., 2005), creep, solifluction and biological and frost weathering (Temme and Veldkamp, 2009) and fluvial behaviour (Baartman et al., 2012a). Furthermore, an algorithm to deal with natural depressions has been added (Temme et al., 2006) and event-based modelling with an improved infiltration description has been pursued (Buis and Veldkamp, 2008). Detailed descriptions of LAPSUS can be found in Chapter 2, 4 and 5. For this thesis, LAPSUS has been enhanced to incorporate separate erodibility values for a 3D lava dam body (Chapter 2) and redistributed sediments (Chapter 4 and 5). Furthermore, the routine that deals with deposition in depressions has been enhanced to be able to deal with multiple outlets (Chapter 2).
1.5 Study area
Research was done in the Upper Gediz River stretch, located north of Kula (Manisa Province, western Turkey, Fig. 1.1). The Gediz River has formed a gorge in an uplifting footwall block of the Alaşehir graben and has cross-cut and incised into Miocene alluvial and lacustrine interior basin deposits, and it has locally exhumed meta-sedimentary basement rocks (Maddy et al., 2007). The area comprises a part of the Kula alkali volcanic field. In the first century, the Greek geographer Strabo visited the area, travelling upstream the Hermus River (the current Gediz) and described it as “Catacecaumene”, which means “burnt land”. In his translated description the area is described as follows: “The surface of the plains is covered with ashes, but the hilly and rocky part is black, as if it were the effect of combustion” (Falconer, 1903). However no historical account of active volcanism in the Kula area is known. Previous work showed that volcanism repeatedly produced lava flows which entered, filled and sometimes dammed the Gediz River in the early Pleistocene (Maddy et al., 2007; Maddy et al., 2012a), middle Pleistocene and Holocene (Bunbury et al., 2001) at consecutively lower levels due to river incision (Fig. 1.2). These levels were named β2, β3 and β4, respectively (Richardson-Bunbury, 1996). This caused the formerly valley-filling basalt to remain as plateaus and ridges, creating a stepped relief-inverted landscape. Incision by the Gediz since 1 Ma has been around 140 m, and tributaries had to respond to this incision. The confluence of a small tributary, the Geren, with the Gediz River is located in the area were multiple lava flows are observed along the present river bed, suggesting that this tributary has been dammed by these lava flows. The Gediz stretch upstream of this confluence and the Geren Catchment were the main study sites in this research. The current landscape of the Geren consists of a ridge-gully landscape, where gently sloping palaeo surfaces are steeply dissected by gullies, sometimes leading to a badland morphology. The altitude of the study area ranges from around 300 to 850 m. Current average annual precipitation is around 600 mm with dry summers and wet winters. Average annual temperature is 15 °C, with hot summers and cold winters with a possibility of snowfall.
1.6 Aim and research questions
Several studies thus emphasize the importance of local history of a catchment on Quaternary timescales, and the influence past events can have on current catchment evolution (Mather et al., 2002). Past lava damming events are a good example of such a past event as they can be long-lived and can exert a long-term influence on catchment evolution (Ely et al., 2012; Maddy et al., 2012a). The Quaternary evolution of the Geren Catchment has been influenced by lava damming events which influenced its base level. However, catchment response is unknown. Field-based landscape reconstruction and landscape evolution modelling are
two methodologies to approach this problem (cf. Temme, 2008). These two approaches are
combined to unravel landscape evolution on a 105 ka timescale.
In this thesis, both inductive, explorative fieldwork-based landscape reconstruction and deductive, geomorphic-transport-law-based landscape evolution modelling are conducted and combined to understand landscape evolution since the middle Pleistocene of a Mediterranean upland catchment (the Geren Catchment) that has endured multiple lava damming events. The following sub-objectives and research questions were formulated: 1. Simulate the impact of natural damming on upland catchment evolution.
• What is the impact of damming on net erosion?
• How are planform stream routing and longitudinal stream profiles affected through time?
• Is this response influenced by dam and landscape substrate, and if so, how? • Do these results relate to field situations and how?
2. Investigate the impact of Holocene lava damming and breaching on adjacent Gediz and Geren reaches.
• What was the age and duration of the Holocene damming event? • What was Gediz River and Geren Catchment response to damming?
• How did breaching occur and what was the Gediz River and Geren Catchment response?
3. Simulate the long-term response of a small tributary catchment on multiple damming events in relation to 300 ka of gradual base level lowering.
• What is the effect on net erosion?
• What is the effect on longitudinal profile evolution?
• What is the effect on spatial patterns of net erosion and sediment storage?
4. Reveal lava dam-influenced base level evolution of the Geren Catchment since the middle Pleistocene and unravel Geren Catchment response to this evolution.
• Can palaeo-Gediz levels and lava damming events be identified and age-constraint? • Can palaeo-Geren levels be identified and age-constraint?
• How did base level at the Geren Catchment outlet evolve since the identified damming events?
• What was the Geren Catchment evolution, based on fieldwork evidence?
• What is the Geren Catchment evolution, based on its base level evolution and landscape evolution modelling?
1.7 Thesis outline
The scope of this thesis spans an array of fieldwork and modelling approaches (Fig. 1.3). Chapter 2 is a model study based on an artificial catchment and presents results which demonstrate different responses to differences in dam and substrate erodibility. Chapter 3 contains fieldwork results only, leading to a conceptual diagram of dam response. Chapter 4 is a modelling study, already closer to our field situation and where response of a catchment is modelled by changing uplift and damming as external drivers. Chapter 5 shows an integrated approach, where fieldwork yields chronostratigraphical control and some initial conclusions. Modelling subsequently uses these results and gives insight in potential mechanisms of chronostratigraphy. Another way to subdivide Chapters is their applicability. Chapter 2 is the least related to a specific field situation, and its conclusions can be taken to be the most conceptual, followed by Chapter 4. Chapter 3 and 5 are the most specific to the fieldwork area.
Landscape evolution modelling of naturally dammed rivers
Natural damming of upland river systems, such as landslide or lava damming, occurs worldwide. Many dams fail shortly after their creation, while other dams are long-lived and therefore have a long-term impact on fluvial and landscape evolution. This long-term impact is still poorly understood and landscape evolution modelling can increase our understanding of different aspects of this response. Our objective was to simulate fluvial response to damming, by monitoring sediment redistribution and river profile evolution for a range of geomorphic settings. We used landscape evolution model LAPSUS, which calculates runoff erosion and deposition and can deal with non-spurious sinks, such as dam-impounded areas. Because fluvial dynamics under detachment-limited and transport-limited conditions are different, we mimicked these conditions using low and high erodibility settings, respectively. To compare the relative impact of different dam types, we evaluated five scenarios for each landscape condition: one scenario without a dam and four scenarios with dams of increasing erodibility. Results showed that dam-related sediment storage persisted at least until 15000 yr for all dam scenarios. Incision and knickpoint retreat occurred faster in the detachment-limited landscape than in the transport-limited landscape. Furthermore, in the transport-limited landscape, knickpoint persistence decreased with increasing dam erodibility. Stream capture occurred only in the transport-limited landscape due to a persisting floodplain behind the dam and headward erosion of adjacent channels. Changes in sediment yield variation due to stream captures did occur but cannot be distinguished from other changes in variation of sediment yield. Comparison of the model results with field examples indicates that the model reproduces several key phenomena of damming response in both transport-limited and detachment-limited landscapes. We conclude that a damming event which occurred 15000 yr ago can influence present-day sediment yield, profile evolution and stream patterns.
Published as: Van Gorp, W., Temme, A.J., Baartman, J.E., Schoorl, J.M., 2014. Landscape Evolution Modelling of naturally dammed rivers. Earth Surface Processes and Landforms.
Natural damming of upland river systems occurs worldwide (Costa and Schuster, 1988; Korup, 2002). Dams can be formed by landslides, volcanic edifices and lava flows, ice and glacial
landforms such as moraines (Costa and Schuster, 1988; Korup and Tweed, 2007; O’Connoret
al., 2013), fluvial activity such as levees or alluvial fans and bio- and hydrochemical systems
such as travertine (Florsheimet al., 2013). Landslide dams are the most common dam type,
occurring worldwide in high relief areas (Costa and Schuster, 1988). Landslides mostly are triggered by tectonic activity, intense rainfall or snowmelt events, as well as volcanic activity (Capra, 2007). Landslide dams are often short-lived and fail catastrophically within weeks (Costa and Schuster, 1988; Walder and O’Connor, 1997; Ermini and Casagli, 2003). However, longer-lived dams do occur and these can have a prolonged effect on fluvial evolution (Korup et al., 2006, 2013). Landslide dam stability largely depends on dam height, volume, material properties and the water influx into the dammed area (Costa and Schuster, 1988; Ermini and Casagli, 2003). Small dams made of loose matrix-supported debris flow material blocking a large catchment will be short lived, while large dams consisting of large blocks have longer lifespans (e.g. Capra, 2007). Other dams that generally have higher longevity are lava dams. These are not as widespread as landslide dams, but they can have significant impact on local or regional catchment evolution due to their relatively high longevity. Examples of lava dams and their effects are found along the Allier River in France (Macaireet al., 1992), the Tana River in Kenya (Veldkampet al., 2007; 2012), the Gediz River in Turkey (Maddyet al., 2012a; Van Gorpet al., 2013) and the Colorado River (e.g. Hamblin, 1990; Dalrymple and Hamblin, 1998; Fentonet al., 2004, 2006; Crowet al., 2008) and Owyhee River in the U.S.A (Elyet al., 2012). A conceptual model of lava dam types presented in Crow et al. (2008) indicates the relatively long-lived nature of massive lava dams against the short-lived nature of more permeable dams. It must be noted however that if dam seepage equals inflow in the impoundment, lake filling does not occur, hampering dam destruction. The result of damming can be either partial or complete blocking of a river valley. For landslide dams, several types of blocking have been defined (Costa and Schuster, 1988), which can be extended to other dam types such as glacial or lava dams. For instance, a dam can fill a valley along its axis and block tributary valleys as well, block the same river at multiple locations or occur as a combination of these examples. The response of the fluvial system to damming can include rapid dam removal, lake formation, drainage rearrangement, backwater aggradation and lake siltation. The dam can become a temporary local base level for a river expressed as a knickpoint in the river profile. These
knickpoints can persist due to bed armouring by large blocks (Korupet al., 2006) and in
mountainous areas, they are known to last up to 104 a (Korupet al., 2006, 2010). Persistence of sediment wedges could create hillslope-channel decoupling due to more complex landscape morphology (e.g. Baartmanet al., 2013a), while erosion of these sediments after dam removal can create downstream aggradation, unpredictable sediment yields and river profile evolution. So far, analysis of catchment response to dams of various size and origin has mainly been done by field-based reconstruction studies (e.g. Korupet al., 2006; García-Garcíaet al., 2011). These studies are valuable, but often restricted to specific cases, limiting wider applicability. Damming also emerged as a controlling factor in some Quaternary catchment reconstruction studies (e.g. Baartmanet al., 2011; Maddyet al., 2012a; Veldkampet al., 2012; Van Gorp
Despite these advances, there is insufficient understanding of which response follows which type of damming. Landscape evolution models (LEMs) can perform structured experiments that may provide such understanding. LEMs are nowadays widely used to understand and quantify landscape response to driving factors or to simulate internal complexity and process interactions. Several studies demonstrated complex landforms emerging from relatively simple drivers or process interactions. Examples are model studies on river-vegetation interaction (Murray and Paola, 2003), hillslope-river coupling for experimental catchments (Baartman et al., 2012a) and interaction of several mass movement processes and water flow interaction for an actual landscape (Temme and Veldkamp, 2009). At the same time, landscape evolution modelling still faces challenges to improve functionality and areas of application (Tucker
and Hancock, 2010; Temmeet al., 2011a). Landscape evolution modelling has not yet been
applied to simulate response to damming. To simulate this response, two important aspects need to be incorporated in a LEM: adequate depression filling routines to deal with creating and filling depressions and incorporation of lithological boundaries between dams and the underlying substrate.
Since dammed landscapes involve natural depressions (e.g. lake formation), modelling these landscapes should include a routine to naturally deal with depressions. Water flow has to be routed to the depression outlet and sediments entering the depression should be deposited below water level. Several LEMs have a routine to deal with depressions dynamically (e.g.
CHILD (Tucker et al., 2001), CEASAR (Coulthard et al., 1998) and LAPSUS (Temme
et al., 2006)). The routine in LAPSUS is arguably most advanced because it allows for the
formation of simplified deltas in depressions and for the fragmentation of depressions. With these routines, researchers have used LEMs to study the impact of depressions on catchment evolution (Hancock, 2008; Temmeet al., 2011a).
It is our objective to provide structural understanding of catchment response to long-lived dams with a LEM that has been adapted for this purpose. Our main research questions are:
• What is the impact of long-lived dams on sediment redistribution? • What is the impact of long-lived-dams on river profile evolution? • How do different dam and substrate erodibilities influence these results?
To achieve our objective, idealised dams in a small, idealised catchment are used in simulations of catchment response.
We selected LEM LAPSUS for this study, due to its advanced sink-filling routines, which will be discussed below. For further discussion on different LEMs the reader is referred to recent reviews by Tucker and Hancock, (2010) and Temme et al., (2013). LAPSUS is a cellular automaton model that simulates runoff erosion and deposition (Schoorl et al., 2000, 2002), landsliding (Claessens et al., 2005, 2007), weathering, creep and solifluction (Temme and Veldkamp, 2009) and which can mimic fluvial behaviour (Baartman et al., 2012a). To limit
the added effect of process interaction and complexity, we only used the runoff erosion and deposition process. LAPSUS deals with non-spurious sinks, which either could be small sinks in river channels or large lake-sized sinks (Temme et al., 2006). This last capability is particularly interesting for this study and is generally lacking in other LEMs. Below, the different model routines that are important for this study are discussed.
2.2.2 Water erosion and deposition
Water and sediment are routed down from each cell to its downstream neighbours using the
multiple flow algorithm (Freeman, 1991; Quinn et al., 1991). Sediment transport capacity C
(m) over time t (yr) and space s (m) between two cells is then calculated from the fractional discharge Q (m) and tangent of slope Λ (-) (Kirkby, 1971):
Cs,t = Qs,t m . Λ
s,t n (2.1)
Parameters m and n are the discharge and slope exponent, respectively (Kirkby, 1987). The amount of sediment that will be transported is then calculated using (Foster and Meyer, 1972, 1975):
Ss,t = Cs,t + (S0 s,t - Cs,t ) . e -cellsize / h (2.2)
where sediment in transport S (m) over one cellsize length is a function of transport capacity
C and erodibility or sedimentation factor h (m), compared with the amount of sediment
already in transport S0 (m). If there is more sediment in transport than the transport capacity, deposition will occur and sedimentation factor h is calculated as follows:
Ps,t . Qs,t . Λs,t hs,t = Cs,t
where P (m-1) is a sedimentation factor. If sediment already in transport is smaller than transport capacity, erosion will occur and h is calculated as follows:
Ks,t . Qs,t . Λs,t hs,t = Cs,t
where K (m-1) is an erodibility factor. Both K and P are lumped factors, representing surface characteristics of a gridcell. This parameter is not an empirical value such as USLE based K-factors. However, it determines the detachment capacity of a cell (Schoorl and Veldkamp, 2001) and has been used as a calibration factor in many previous studies with LAPSUS, both in studies of actual landscapes (Schoorl et al., 2002, 2004; Temme et al., 2009; Temme and Veldkamp, 2009; Baartman et al., 2012b) as well as in experimental studies (Baartman et al.,
2012a). Factors K and P can have different values (Temme and Veldkamp, 2009), however,
they are kept equal in this study for simplicity. A combination of a low K and P implies that the substrate is hard to erode, but as soon as sediment is in transport, it is hard to deposit again.
This mimics detachment limited conditions. On the other hand, a combination of a high K
and P implies that sediment is easy to erode but also easy to deposit, mimicking transport-limited conditions. Furthermore, since sedimentation depends on the amount of sediment in transport, if the erodibility is low, a high sedimentation factor does not necessarily mean high
sedimentation. In this study, K and P of the dam body can be set at a different value than the surrounding and underlying substrate.
2.2.3 Sedimentation routine
An important algorithm in LAPSUS is the sedimentation routine. The sediment in transport that can be deposited according to equation (2.2) is not actually deposited in all cases. To avoid the creation of unrealistic spikes, the amount of sediment that is deposited depends on the lowest higher neighbour of the cell considered. If this amount exceeds the elevation difference between the current cell and its lowest higher neighbouring cell, excess sediment in transport will be “smeared” down using a steepest descent smearing routine, until all the sediment in transport has been deposited. If there is still sediment left after the smearing routine has reached the furthest downstream cell (usually the outlet of the catchment), the remaining sediment will be added to the sediment in transport of the original receiving cell, where it can be transported in the conventional way in the next timestep. This routine mimics annual sediment transport distance and is an effective way to avoid spikes, however, it is acknowledged that sediment can be transported unrealistically far within one timestep, especially in flat areas, such as backwaters or deltas. This routine is explained into more detail in Schoorl et al. (submitted).
2.2.4 Depression fill routine
The depression fill routine in LAPSUS is presented and discussed in Temme et al. (2006). In summary, LAPSUS deals with depressions by first defining the extent of all depressions and their outlet(s). Then it collects all the water and sediment flowing into a depression from its contributing area. In this study, it is assumed that a depression is filled with water at the start of the simulation, thus, that the lake is filled up to the outlet elevation. Therefore all water and sediment is routed to and equally divided over the depression’s outlet(s). This capability to deal with multiple outlets was developed for this study and was not yet described in Temme et al. (2006). As long as the depression has not been filled up with sediments, the depression outlets cannot be eroded and catastrophic failure at the outlet of an impounded area also does not occur. The model therefore only removes dam outlets by gradual incision after infilling. In this way, the role of sediment as an abrading tool at the depression outlet is taken into account (Cowie et al., 2008; Ely et al., 2012).
If a depression has a volume smaller than the total incoming sediment, it is completely filled with sediment, to an almost-flat surface, draining to each of the depression’s outlets. The steepness of this surface is user-specified and corresponds to Planchon and Darboux’s (2002) epsilon variable. The water volume that is displaced by sediment in the depression is added to the water flow draining from that depression.
If a depression has a volume larger than the total incoming sediment, sediment is deposited in deltas growing from every side-cell of the depression, with user-specified underwater steepness.
2.2.5 Boundary condition at outlet
The boundary condition used in this study is that the catchment outlet cell, thus the lowest cell at the edge of the DEM, is not considered for erosion or deposition. It therefore will never erode or deposit according to equations (2.1 – 2.4) and thus forms a base level of the catchment. However, sediment can be deposited on this lowest cell due to the sedimentation smearing routine. This potential base level rise can have implications on landscape evolution of the upstream area. We have minimized these implications by extending the experimental catchment significantly downstream of the dam edge.
2.2.6 Experimental design
Because the distinction between landscapes and rivers having detachment-limited and transport-limited conditions is often made (e.g. Tucker, 2009), we chose to study response to damming under these two landscape conditions. Additionally, the erodibility of the dam has been varied in four steps in each landscape to represent weakly and strongly erodible dams. Two additional simulations were done on low-erodibility and high erodibility landscapes without a dam. This resulted in a total of ten scenarios (Table 2.1).
In LAPSUS, the scenarios were translated into changes of the erodibility (K) and sediment
potential (P) values (Table 2.1). The values used are within the range of calibrated K and P
values used in literature (Schoorl et al., 2002, 2004; Temme and Veldkamp, 2009; Baartman et al., 2012a). For all simulations, annual rainfall was 700 mm, with an infiltration loss of 150 mm and a total evaporation loss of 350 mm. This conforms to a Mediterranean climatic setting. All 10 scenarios were run for 15000 yr to be able to record response to damming on a relevant timescale (Korup et al., 2010). Furthermore, landscapes were in a transient (i.e. non-equilibrium) state. The reason for this is that transient conditions may be the norm rather than the exception in landscape evolution (Tucker, 2009). Especially smaller tributary catchments, such as our experimental catchment, are often observed to be in transient condition in reaction to a base level perturbation (e.g. Snyder et al., 2003). Although we acknowledge that differential basin scale transient response might obscure dam perturbations, our starting landscape for both the detachment-limited and transport-limited landscape will be the same, which would not be the case if we first brought both these landscapes into equilibrium conditions.
Table 2.1. Simulated scenarios.
Scenario K and P substrate K and P dam
A: transport limited
A0: no dam 0.0003
A1: K and P dam << K and P substrate 0.0003 0.000003
A2: K and P dam < K and P substrate 0.0003 0.00003
A3: K and P dam = K and P substrate 0.0003 0.0003
A4: K and P dam > K and P substrate 0.0003 0.003
B: detachment limited
B0: no dam 0.00003
B1: K and P dam < K and P substrate 0.00003 0.000003
B2: K and P dam = K and P substrate 0.00003 0.00003
B3: K and P dam > K and P substrate 0.00003 0.0003
B4: K and P dam >> K and P substrate 0.00003 0.003
2.2.7 Experimental input DEM
The artificial input DEM (Fig. 2.1) is based on the DEM designed for the study of Baartman et al. (2012a). It measures 2100 x 6000 m with a resolution of 20 m (31500 cells). Compared to the DEM of Baartman et al. (2012a), slopes and profile gradients were steepened to create a simple, confined valley which is more suitable to model response to damming. It has a profile gradient around 12.5% in the upper part and 1.5% in the lower part and resembles a small upland catchment. The elevation range is 470 m. The dam was designed with two equally high outlets, 14.5 m above the deepest point of the valley at the dam location. One outlet routes over the dam body and the other outlet routes around the dam, where the dam and substrate touch. These locations were chosen to create the opportunity for incision to occur through the substrate or the dam, as both are observed in the field (Ouimet et al., 2008; Korup et al., 2010). The route over the dam drops down once it passes the dam, whereas the route along the dam has a longer pathway and a more gentle slope back towards the valley floor. The equal elevation of both outlets was designed in this way to test the effect of different erodibilities on the development of flow routing.
Fig. 2.1. Left: visualisation of DEM used for model simulations with LAPSUS. Right: Same DEM, dam body is highlighted in pink, areas with highest flow accumulation at t = 1 are added. Two spillways are visible, one over the dam (left) and one around the dam (right).
2.2.8 Evaluation characteristics
The following catchment characteristics were recorded from model simulations for evaluation of results.
1. Lake fill speed
2. Net erosion from the catchment 3. Drainage rearrangement 4. River profile development
5. Occurrence of phenomena derived from different field based and conceptual studies on fluvial response to natural damming.
6. Sediment storage on a cell
To be able to record sediment storage on a cell, a new capacity was added to the model. For each cell the minimum elevation since the start of the simulation is stored and subtracted from the current DEM. Unlike a map of cumulative net change in altitude, this records sedimentation locations and thicknesses even if the cell has experienced net erosion since the start of the simulation.
2.3.1 Annual net erosion
Total annual net erosion varies between 40 and 100 Mg ha-1 for the A-scenarios (high
erodibility, transport-limited) and between 4.0 and 13.0 Mg ha-1 for the B-scenarios (low
erodibility, detachment-limited). For all scenarios, net erosion decreases over time (Fig. 2.2), which is because average slope is declining over time. Annual net erosion values of the A-scenarios are highly variable due to complex behaviour of sediments in transport which are repeatedly eroded and deposited within the catchment. Annual net erosion of the B-scenarios shows less variation, although variation increases for scenarios B3 and B4, which have more erodible dams. Introducing a more erodible surface, in other words introducing transport limited conditions, thus increases the variation of net erosion values. The duration of lake filling by sediments is around 50 yr for the transport-limited A-scenarios and around 400 yr for the detachment-limited B-scenarios (Table 2.2). This is due to the difference in sediment supply to the lake. The cumulative net erosion after 10000 and 15000 yr is always lower for the dam scenarios than for the non-dam scenarios, indicating a consistent long-term sediment storage effect.
This sediment storage effect is different for each dam scenario. Cumulative net erosion values and the amount of sediment stored in the originally inundated dam area are shown in Table 2.2. Total net erosion values of scenarios A3 and B2 (K, P dam = K, P substrate) differ least from their non-dam scenarios, respectively. For the transport-limited A-scenario, net erosion of A2 (K, P dam < K, P substrate) differs the most with A0 (no dam), whereas for the detachment-limited B-scenario, B3 at 10000 yr (K, P dam > K, P substrate) and B4 at 15000 yr (K, P dam >> K, P substrate) differ the most with B0 (no dam). These differences reflect the summed effect of differential evolution of water and sediment routing as a complex response to initial erodibility.
2.3.2 Water and sediment routing evolution
Maps of flow accumulation, cumulative erosion and locations and thickness of sediments are shown for 1000, 5000 and 10000 yr for all dam scenarios and 15000 yr for the A-scenarios only because the B-scenarios at 15000 yr do not provide more information except for additional incision (Fig. 2.3 and Fig. 2.4). Green colours on the map indicate the thickness of redeposited sediment at a certain location. Note that these green colours can also occur on cells with net erosion. Yellow to red colours indicate low to high net erosion in the case that no sediment is present on that cell. In all dam scenarios, most of the catchment surface is only eroding. However, sedimentation is clearly visible upstream of the dammed area. In some cases, water routing over this filled area shows bifurcations and the course of the trunk stream changes over time. Sedimentation is also visible in some of the main channels and sediment storage locations vary over time. All scenarios still have notable dam remains at 15000 yr. In scenario A4 (K, P dam > K, P substrate), the dam is highly erodible and more erodible than the landscape, leaving only a very small part of the dam crest not entirely removed. In all scenarios, a large portion of the sediment initially deposited in the lake remains stored in the landscape at the end of the simulation (See Table 2.2). Scenario A1 even has more sediment
storage in the filled lake at 15000 yr compared with sediment storage at the time of lake filling, indicating net sedimentation since lake filling.
Scenario A1 (K, P dam << K, P substrate) shows the initial establishment of the trunk stream around the dam where it will remain until 15000 yr. Scenario A2 (K, P dam < K, P substrate) shows initial establishment of the trunk stream over the dam. Rerouting of the stream from over the more resistant dam to around the dam over the less resistant substrate occurs just after 10000 yr.
Scenario A3 (K, P dam = K, P substrate) shows initial establishment of the stream around
the dam followed by stream rerouting 15 yr later. At t = 15000 yr, the trunk stream is routed around the dam. However, three stream reroutings have taken place between 7700 and 12250 yr (see Table 2.2). After the last rerouting, two periods of active headward erosion take place (Table 2.2), which coincide with two periods of increased variation of net erosion (Fig. 2.2). Scenario A4 (K, P dam > K, P substrate) also shows initial stream routing over the dam, followed by 4 reroutings in the period from 4000 to 8000 yr. These events do not coincide exactly with periods of increased and decreased variation of net erosion. Detachment-limited scenarios B1 (K, P dam < K, P substrate), B2 (K, P dam = K, P substrate) and B3 (K, P dam > K, P substrate) establish their initial channel around the dam (Fig. 2.4). They subsequently incise to form a gorge and remain in their initial location until the end of the simulation. Scenario B4 (K, P dam >> K, P substrate) establishes its initial channel through the dam and creates a gorge where the channel remains until the end of the simulation.
Fig.2.2. Annual net erosion. Left: A-scenarios, high erodibility and sedimentation factor, representing transport-limited conditions. Arrows indicate the moments of stream captures. Red traces below timeseries line indicate periods of diverging water flow on top of lake fills. Right: B-scenarios, low erodibility and sedimentation factor, representing detachment-limited conditions.
Fig. 2.3. Output maps of transport-limited scenarios A1 – A4 for the reach between 2400 m and 5200 m at different timesteps showing spatial evolution of the catchment reach around the dam. Cumulative erosion (yellow to red), sediment storage (green) and main water routing channels (blue) are depicted. Note distance labels attached to the top-left figure.
Fig. 2.4. Maps of detachment-limited scenarios B1 – B4 for the reach between 2400 m and 5200 m at different timesteps, showing spatial evolution of the catchment reach around the dam. Cumulative erosion (yellow to red: low to high), sediment storage (green) and main water routing channels (blue) at different timesteps are depicted. Note the distance labels attached to the top-center figure.
Table 2.2. Total net erosion and sequence of events.
Total net erosion
(Mg/ha) Lake sediment storage (m3) Event Time (yr) Channel through or around
t=10000 t=10000 t=15000 t lake
filled t = 15000 A: transport-limited
Scenario A0: No dam
22.7 743190 975152
-Scenario A1: K and P of dam 100 times smaller than K and P of substrate
22.8 631859 833043 2.06E+06 2.10E+06 Lake filled t = 44 Around
Scenario A2: K and P of dam 10 times smaller than K and P of substrate
24.2 623764 822341 2.08E+06 1.90E+06 Lake filled t = 64 Through
Capture t = 10305 Around
Scenario A3: K and P of dam equal to K and P of substrate
24.2 662220 875873 2.08E+06 1.18E+06 Lake filled t = 57 Around
Capture t = 67 Through Capture t = 7700 Around Capture t = 9705 Through Capture t = 12239 Around headward erosion t≈12800 Around headward erosion t≈14200 Around
Scenario A4: K and P of dam 10 times larger than K and P of substrate
21.2 648822 852381 2.08E+06 1.05E+06 Lake filled t = 44 Through
Capture t = 4082 Around
Capture t = 7381 Through
Capture t = 7421 Around
Scenario Outlet elevation
Total net erosion
(Mg/ha) Lake sediment storage (m3) Event Time (yr) Channel through or around dam? t=10000 t=10000 t=15000 t lake filled t = 15000 B: detachment-limited Scenario B0: No dam 17.9 123543 182492
-Scenario B1: K and P of dam 10 times smaller than K and P of substrate
17.9 118430 176684 2.05E+06 1.30E+06 Lake filled t = 383 Around
Scenario B2: K and P of dam equal to K and P of substrate
17.9 121000 180218 2.07E+06 1.16E+06 Lake filled t = 385 Around
Scenario B3: K and P of dam 10 times larger than K and P of substrate
20.0 116292 175270 2.05E+06 1.21E+06 Lake filled t = 347 Around
Scenario B4: K and P of dam 100 times larger than K and P of substrate
20.0 120565 172618 1.78E+06 1.14E+06 Lake filled t = 252 Through
2.3.3 Profile evolution
For all scenarios, the longitudinal profile mostly shows net incision over time (Fig. 2.5). Aggradation occurs upstream of the dam and knickpoints are generated at the dam outlet, generally migrating backwards.
The longitudinal profiles of the A-scenarios show less deep incision than the B-scenarios. The zoomed-in section of longitudinal profiles between 2000 m and 5000 m distance along the valley axis (Fig. 2.5), shows profile evolution around the dam location. The A0 scenario (no dam) shows that this part of the profile remains at a constant level after 1000 yr. For the other A-scenarios, trunk stream incision upstream of the dam is more than 20 m less than in the A0 scenario. Scenario A1 (K, P dam << K, P substrate) shows the profile of the route around the dam. At the dam location, a set of stepped knickpoints is created which remain quite stable throughout the 15000 yr. Scenario A2 and A3 show 1.5 to 2 m higher elevation at the outlet than A0 (Table 2.2), which is due to higher aggradation. Scenario A2 (K, P dam < K, P substrate) shows a steep drop near the resistant dam spillway until 10000 yr. At 15000 yr, which is 5000 yr after stream capture occurred, the knickpoint has become less steep and
has migrated backwards. Scenario A3 (K, P dam = K, P substrate) shows modest backwards
incision from 1000 to 5000 yr. Between 5000 and 10000 yr, two captures occur. Despite this capture, hardly any backward knickpoint migration occurred. At 15000 yr a deeper incised profile with several smaller knickpoints is visible. Scenario A4 (K, P dam > K, P substrate)
shows a backwards migrating knickpoint that becomes less pronounced. Captures take place between the profiles of 1000 and 5000 yr, and 5000 and 10000 yr, thus the profiles do not show straightforward backwards incision. Between 5000 yr and 10000 yr, backwards knickpoint migration is nearly zero, despite the fact that the trunk stream position has changed.
For detachment-limited scenarios B1, B2, B3 and B4 a knickpoint is visible at 1000 yr. Profiles of B1 (K, P dam < K, P substrate) and B2 (K, P dam = K, P substrate) at 5000 and 10000 yr are similar to those from scenario B0 (no dam), indicating that the effect of the dam on profile evolution has diminished over time due to fast headward knickpoint migration. Scenario B3 (K, P dam > K, P substrate) shows moderate backfilling in the trunk stream after incision. Scenario B4 (K, P dam >> K, P substrate) shows enhanced backfilling at 10000 yr and 15000 yr, after initial incision. This can be attributed to higher dam erodibility and higher sediment supply to the trunk stream from dam slopes and indicates a long-term effect of locally high sediment supply to profile evolution.
Fig. 2.5. Longitudinal profile evolution for all scenarios for the downstream reach of the catchment for different timesteps. X-axis shows the distance along the valley-axis (from 2400 to 6000 m). The profile of t = 1 yr shows the location and altitude of the dam. Arrows indicate direction and rate of knickpoint retreat. For scenarios B0 – B3, the longitudinal profile of t = 15000 yr is not shown because this profile does not differ significantly from that on t = 10000 yr.
2.4.1 Scenario characteristics
Some differences were observed between the A- and B-scenarios irrespective of dam presence or absence. The transport-limited setting of the A-scenarios generates higher erosion on the slopes resulting in high sediment loads in the trunk stream. This can correspond to landscapes on highly erodible bedrock such as marls with low-vegetation cover, such as Mediterranean landscapes or degraded or abandoned agricultural landscapes. The transport-limitation is expressed in higher net erosion, higher sedimentation, lower incision and a convex shape of the trunk stream (Tucker, 2009). The transport-limited setting is also expressed in the complex variation of the net erosion timeseries. This variation has been observed in other landscape evolution modelling studies, and are similar to the nonlinear temporal dynamics illustrated by Van De Wiel and Coulthard (2010). The spatial expression of this variation is reflected in variations of sediment storage within streams over time (Fig. 2.3). Within the A-scenarios, the presence of a dam increases variation of net erosion, irrespective of dam erodibility. This points to a more complex transfer of sediment from behind the dam to the catchment outlet, which can be attributed to the more complex sediment transfer over the nearly flat surface space behind the dam. Total net erosion of the A-scenarios does not increase linearly with increasing dam erodibility (Table 2.2). For instance, despite the fact that scenario A1 has the least erodible dam and thus the most potential to store sediments, scenario A2 has the lowest total net erosion at 10000 yr and 15000 yr. This is determined by the specific history of the scenario. Scenario A1 establishes a route around the dam on top of more erodible substrate and remains there, while scenario A2 establishes a route over the less erodible dam, leading to more storage behind the dam. In both scenarios, incision in the first 10000 yr is limited. However, the trunk stream of scenario A2 is captured by a headward eroding stream at 10305 yr. Nevertheless, total net erosion remains higher in scenario A1, illustrating how this signal reflects the summed effect of different internal evolution events.
The detachment limited setting of the B-scenarios generates less erosion on the slopes resulting in low sediment loads in the trunk stream. This is expressed in lower net erosion, low variation of erosion over time, low sedimentation and deep incision of the trunk stream. This deep incision is due to sufficiently high water flow with low amounts of sediment in transport and thus high unfulfilled transport capacity. This also explains the low variation of net erosion: the bulk of the material is directly eroded from the stream bed of the trunk streams and immediately removed. A large part of the landscape as a sediment source is disconnected from the trunk stream, and the effective catchment area is small. This limits the influence of landscape complexity on net erosion to the trunk gullies only. Geomorphologically, this corresponds to landscapes found either on resistant bedrock, or under dense vegetation. Although response to damming in both the A- and B-scenarios occurs through lake filling and subsequent re-incision, the high sediment load of the trunk stream in the transport-limited A-scenarios causes lake filling to occur faster. Low sediment load of the trunk stream causes backwards knickpoint migration to occur faster in the detachment-limited B-scenarios. Stream rerouting does occur in the A-scenarios and does not occur in the B-scenarios. The reason for this is that in all of the B-scenarios, incision into the substrate
and headward erosion is occurring at such a high rate that incision of smaller streams cannot keep up. In the A-scenarios, incision by the trunk stream is less severe. A probable reason for this slow migration is the low slope of the trunk stream on top of the lake sediments, combined with a high sediment load of the trunk stream, reducing eroding capacity. At the knickpoint, erosion is hampered due to the high sediment load, despite the steep slope. The gradual diminishing of the knickpoint suggests that the fluvial profile of the transport-limited stream is gradually changing into its linear to convex shape, similar to the profiles of A0 (and also see Tucker, 2009). The knickpoint remains the most pronounced for scenario A1 and A2 (K, P dam < K, P substrate) and least pronounced for A4 (K, P dam > K, P substrate). The lack of gorge formation makes potential capture by a smaller stream more likely. This suggests that in landscapes with more easily eroded lithology, stream capture is more likely to accompany damming than in landscapes with less erodible lithology.
2.4.2 Drainage rearrangements
Drainage rearrangement can be caused by aggradation processes, leading to avulsions, or by headward erosion driven stream capture. The frequent bifurcations of the trunk stream on top of the filled lake (Fig. 2.3 and Fig. 2.4) suggest that sediment redistribution on this floodplain plays a role, causing avulsions and potential switch of drainage from one outlet to the other. However, in all scenarios experiencing rearrangements (Table 2.2), the trunk stream at the dam location is flanked by a smaller stream discharging into the other outlet. The smaller stream manages to erode headwards faster than the trunk stream, despite that catchment area and thus water flow of the trunk stream is significantly larger than water flow of the smaller stream. For scenario A2 (K, P dam < K, P substrate), this stream capture can be explained by a higher erodibility of the substrate below the smaller stream around the dam (Fig. 2.6). For scenario A3 (K, P dam = K, P substrate) and A4 (K, P dam > K, P substrate), a possible cause for this behaviour is the high sediment load of the trunk stream in comparison with the smaller stream. Thus, the trunk stream is locally behaving like a transport limited stream, while the smaller stream is locally behaving like a detachment limited stream. This is visible by backwards knickpoint migration in the small stream after every capture and resembles the behaviour of the conceptual diagram depicted in Schumm (1979, Figure 15), where transport-limitation and detachment-limitation drive ongoing stream capture. The observed drainage rearrangements in our simulation generally start with a stream branching off the trunk stream into the smaller stream, with lower runoff than the trunk stream. Subsequently, rerouting of the trunk stream takes place and the main discharge switches either immediately or after a variable amount of time (up to more than 100 yr). After rearrangement, several years of continuing divergence in the streamflow of the two main streams can occur. The exact time and location of connection from the trunk stream to the smaller stream is not directly driven by the stream committing the piracy. Rather, local sediment redistribution and its influence on local small-scale topography determines this exact time and location. A change in net erosion variation after a capture is usually observed (Fig. 2.2), although comparable changes in variation can be observed when there is no preceding drainage rearrangement. This is an indication of unpredictable catchment behaviour, which has been investigated more systematically by Van de Wiel and Coulthard (2010). Thus, although capturing does affect the nature of sediment yield, captures cannot be distinguished from the catchment wide net erosion signal.
Fig. 2.6. Water routing (top) and profiles (bottom) before and after a river capture event of scenario A2. Profile at t = 1 is also displayed. Arrows added to maps indicate main drainage direction.